viva_glyph - v1.0.0

viva_glyph/encoder

Encoder - PAD to Latent Space transformation

Expands 3D PAD state into 6D latent space optimized for emotional quantization (Johnson-Lindenstrauss optimal).

Architecture

PAD (3D) ──▶ Expand ──▶ Latent (6D) ──▶ RVQ ──▶ Glyph

Expansion includes:
- Raw PAD values (P, A, D)
- Derived features (intensity, valence_sign, activation)

Validated (DeepSeek R1 671B - 2025-01-24)

Types

Context

Context for encoding (optional metadata)

pub type Context {
  Context(situation: Int, temporal: Float)
}

Constructors 

  • Context(situation: Int, temporal: Float)

    Arguments

    situation

    Situation identifier (0 = no context)

    temporal

    Time-based feature (0.0-1.0, e.g., circadian)

EncoderConfig

Encoder configuration

pub type EncoderConfig {
  EncoderConfig(latent_dim: Int, use_derived: Bool)
}

Constructors 

  • EncoderConfig(latent_dim: Int, use_derived: Bool)

    Arguments

    latent_dim

    Latent space dimension (default 6, optimal per Johnson-Lindenstrauss)

    use_derived

    Whether to include derived features

Pad

PAD state (matches viva_emotion)

pub type Pad {
  Pad(pleasure: Float, arousal: Float, dominance: Float)
}

Constructors 

  • Pad(pleasure: Float, arousal: Float, dominance: Float)

Values

decode 

pub fn decode(rvq: rvq.RvqEncoder, glyph: glyph.Glyph) -> Pad

Full decoding pipeline: Glyph → PAD

decode_batch 

pub fn decode_batch(
  rvq: rvq.RvqEncoder,
  glyphs: List(glyph.Glyph),
) -> List(Pad)

Batch decode multiple glyphs

default_config 

pub fn default_config() -> EncoderConfig

Default encoder config

encode 

pub fn encode(
  rvq: rvq.RvqEncoder,
  pad: Pad,
  context: Context,
) -> glyph.Glyph

Full encoding pipeline: PAD → Glyph

encode_batch 

pub fn encode_batch(
  rvq: rvq.RvqEncoder,
  pads: List(Pad),
  context: Context,
) -> List(glyph.Glyph)

Batch encode multiple PAD states

latent_to_pad 

pub fn latent_to_pad(latent: List(Float)) -> Pad

Contract latent vector back to PAD Only uses first 3 components (ignores derived features)

no_context 

pub fn no_context() -> Context

No context

pad_to_latent 

pub fn pad_to_latent(pad: Pad, context: Context) -> List(Float)

Expand PAD to latent vector

Latent layout (6D) - validated by DeepSeek R1: [0] pleasure [1] arousal [2] dominance [3] intensity = sqrt(P² + A² + D²) / sqrt(3) - magnitude metric [4] valence_sign = sign(P) * |P|^0.5 - emphasizes valence direction [5] activation = A × D - preserves dominance sign (Yik 2011)

Note: temporal/reserved removed per Johnson-Lindenstrauss optimization

round_trip 

pub fn round_trip(
  rvq: rvq.RvqEncoder,
  pad: Pad,
  context: Context,
) -> #(Pad, Float)

Round-trip with error measurement

✨ Search Document